🔍 Read the full analysis: Muse Spark Open Offers An Alternative To Meta’s Closed Model At A Similar Price on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Meta released Muse Spark Open on October 15, 2026, publishing its 816-billion-parameter model weights on Hugging Face. Artificial Analysis scored it 38.9, with leading results among open-weight models on hallucination and visual reasoning tests, but its token prices match Meta’s higher-scoring closed Muse Spark 1.3. The license has conditions, and the model’s performance on a demanding terminal-agent benchmark was weak.
Meta released Muse Spark Open on October 15, publishing the weights for an 816-billion-parameter model on Hugging Face. Artificial Analysis scored it 38.9 on its Intelligence Index, while the median listed API price matches Meta’s closed Muse Spark 1.3, which scored higher; the open model’s license also comes with conditions.
Muse Spark Open is a mixture-of-experts model with 42 billion active parameters per token and 256 experts, eight of which are active for each token. It accepts text, images, speech and video, produces text, and has a one-million-token context window. Meta published BF16 and NVFP4 weights. Its model card estimates self-hosting requirements of about 700 GB of GPU memory for NVFP4 and 2 TB for BF16, and says it is optimized to run on two Mac Studios with M5 Ultra using MLX.
Artificial Analysis tested the model on release day at its default xhigh reasoning effort. Its Intelligence Index v4.3.2 score was 38.9, rounded to 39. The source report describes that as the highest score for an open-weight model from a US lab and tenth overall among open-weight models; all nine models ahead of it are from Chinese labs. The comparison is specific to that index and its tested entries.
Across four third-party providers, the median list price was $1.25 per million input tokens, $0.15 per million cached input tokens and $4.25 per million output tokens. The source report says those rates are the same as for Meta’s closed Muse Spark 1.3. Artificial Analysis scored that model 45.1 at the same reasoning effort. The weights can be downloaded, but using a provider’s API incurs per-token charges.
Infographic 01 · Where Muse Spark Open lands
The best US open-weight model is tenth. The nine ahead are Chinese.
Artificial Analysis scores Meta’s first open Muse Spark model at 38.9. That is 12 points clear of any other US open-weight model, but behind nine Chinese models. Four of those nine are MIT-licensed, and three of them cost a third or less per task.
Top open-weight models by Intelligence Index
Higher is better. Effort variants of the same model are not listed twice. Cost = average API cost to run one index task.
License: Muse Spark Open ships under the Muse Model License: broad use and redistribution, but mandatory “Built with Muse Spark Open” attribution, a separate agreement for model-as-a-service businesses above $20M revenue, and a usage policy.
Source: Artificial Analysis, Muse Spark Open (xhigh) model page and model data, read 16 Oct 2026. Intelligence Index v4.3.2, ten evaluations. License categories as listed by Artificial Analysis. Countries: Xiaomi, Z AI, StepFun, Kimi, InclusionAI, Alibaba and DeepSeek are Chinese labs; Meta and NVIDIA are US.
Powered by Thorsten Meyer AI
Open Weights, Closed-Model Rates
The release gives developers access to a large model from Meta’s flagship Muse Spark line without requiring access through Meta’s closed model alone. That may matter to teams evaluating self-hosting, adapting weights or building workflows around long documents and mixed media. But the published self-hosting estimates—700 GB of GPU memory for NVFP4 or 2 TB for BF16—make local operation a substantial infrastructure commitment.
For users relying on hosted APIs, the price comparison is less favorable against several open-weight alternatives. Artificial Analysis estimates a weighted cost of $0.84 per Intelligence Index task for Muse Spark Open. The source report puts MiMo-V2.6-Pro at $0.13 for a score of 46.3 and DeepSeek V4.1 Flash at $0.27 for a score of 39.5. Those figures come from the report’s benchmark comparison; they do not establish the cost of every real-world workload.
Performance also varies by task. Artificial Analysis reported a 24% hallucination rate, the lowest among the 15 highest-scoring open-weight models in its comparison, and an 81% score on MMMU-Pro visual reasoning, the top result among 66 open-weight entries tested. On Terminal-Bench 4.0, however, it scored 8%, well below the closed Muse Spark 1.3 at 33%. That split matters to buyers choosing between research and document work and autonomous coding tasks.
high-performance AI model hosting server
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Meta’s First Open Muse Spark
Muse Spark Open is the first open-weight release from Meta’s flagship Muse Spark line. Meta’s recent open models were smaller: the source report says Muse Glimmer, a 30-billion-parameter model released in August 2026 under Apache 2.0, scored 17 on the current Artificial Analysis index. Llama 4 Maverick, released in April 2025, scored 9. Those comparisons place the new release within Meta’s own open-model lineup, though scores depend on the benchmark version and testing setup.
The release also adds a large US-developed model to an open-weight field where the source report says Chinese labs occupy the top nine places on the index. It cites Xiaomi’s MIT-licensed MiMo-V2.6-Pro at 46.3 as the overall leader. NVIDIA’s Nemotron 3.5 Super VL, released on the same day as Muse Spark Open, scored 26.5. The report characterizes it as the leading US open-weight model among permissively licensed offerings; Muse Spark Open scored higher, but its license has conditions.
The model’s strong results are not uniform across benchmarks. In addition to the results above, Artificial Analysis reported 89% on GPQA Diamond, 44% on Humanity’s Last Exam and 58% on SciCode. For agentic and computer-use evaluations, the source report lists 89.2 on MCP-Atlas, 80.4 on OSWorld Verified and 76.4% on Terminal-Bench 2.1. Its lower Terminal-Bench 4.0 result illustrates why benchmark choice matters when judging a model for coding or agent use.
As an affiliate, we earn on qualifying purchases.
License Terms and Real-World Costs
The available source material does not spell out the specific conditions in Muse Spark Open’s license, so readers should consult the license itself before using the weights in a commercial product or redistributing them. The report distinguishes it from permissively licensed competitors but does not detail which uses are restricted.
API figures are median list prices across four third-party providers, not a guarantee of a user’s final bill. Actual cost depends on provider rates, cached and uncached input, output volume and workload. The source report says the benchmark run used 2.96 billion input tokens across the index tasks, but gives no universal usage pattern that would predict a particular deployment’s spending.
Benchmark scores also do not settle how well the model will perform in a specific product or under different prompts and settings. The source material does not provide a full account of evaluation uncertainty, provider implementation differences or the practical self-hosting performance behind Meta’s hardware estimates.
As an affiliate, we earn on qualifying purchases.
Check Terms Before Deployment
Developers considering Muse Spark Open can review the Hugging Face weights and Meta model card for the license, supported formats and deployment guidance. Hosted users can compare providers’ current rates and test their own input and output patterns against alternatives, since the reported figures are list-price and benchmark comparisons.
For teams considering coding agents, the key next step is to evaluate the model on their own terminal and tool-use tasks. The available results differ sharply between Terminal-Bench 2.1 and the harder Terminal-Bench 4.0 evaluation, so the release-day index alone cannot determine whether it fits a particular workflow.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Muse Spark Open?
It is Meta’s first open-weight model in its Muse Spark line, released October 15, 2026. It has 816 billion total parameters, 42 billion active per token and a one-million-token context window.
How did it score against other open-weight models?
Artificial Analysis scored it 38.9 on its Intelligence Index v4.3.2, rounded to 39. The source report ranks it tenth overall among open-weight models tested and highest among those from a US lab.
Is the API cheaper than Meta’s closed Muse Spark 1.3?
No, according to the source report: the median listed per-token rates across four providers match those for closed Muse Spark 1.3. That model scored 45.1 on the same index at the same xhigh reasoning effort, compared with 38.9 for Muse Spark Open.
Can developers use the weights without restrictions?
The weights are available on Hugging Face, but the source report says the license has conditions. It does not list those conditions, so developers should read the license and model card before deciding how to use or distribute the model.
Is Muse Spark Open suitable for coding agents?
Its results are mixed. Artificial Analysis reported 76% on Terminal-Bench 2.1 but 8% on Terminal-Bench 4.0; the source report also lists weaker results on some other agentic comparisons. Teams should test the model on their own tools and tasks.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
